6716568 fpsd:fpsd daemon gives "kstat_lookup failed" error when processor state is changed to "no-intr"

3 files changed, 41 insertions, 16 deletions

diff --git a/usr/src/cmd/fps/fpsd/fpsd.h b/usr/src/cmd/fps/fpsd/fpsd.h
index 1819023041..679cc8f832 100644
--- a/usr/src/cmd/fps/fpsd/fpsd.h
+++ b/usr/src/cmd/fps/fpsd/fpsd.h
@@ -88,6 +88,7 @@ typedef struct
 	char	m_machine[MAXNAMELEN];	/* machine name e.g. sun4u */
 	uint_t	m_num_fpus;	/* num of fpus in the system */
 	uint_t	m_num_on_fpuids;	/* num of online cpus */
+	int	m_max_cpuid;	/* maximum cpuid for this system */
 	fps_cpu_t	*m_cpus;	/* array of cpus to test */
 	int		m_cpuids_size;	/* size of previous array */
 	int	m_num_cpus_to_test;	/* Num cpus to run test */
diff --git a/usr/src/cmd/fps/fpsd/fpsd_main.c b/usr/src/cmd/fps/fpsd/fpsd_main.c
index 7f54b01d41..b206cac7c8 100644
--- a/usr/src/cmd/fps/fpsd/fpsd_main.c
+++ b/usr/src/cmd/fps/fpsd/fpsd_main.c
@@ -79,7 +79,6 @@ void terminate_process();
 
 static int  door_id = -1;
 static char *str_fps_fmri = NULL;
-static int  max_cpuids = 0;
 
 /* Local static functions */
 
@@ -409,21 +408,19 @@ fpsd_probe(mach_conf_t *m_stat)
 
 	processorid_t *cpuid_list;
 	kid_t ret;
+	int total_onln = sysconf(_SC_NPROCESSORS_ONLN);
 
 	/* probe the system and fill in mach_conf_t elements */
 
 	(void) sysinfo(SI_MACHINE, m_stat->m_machine,
 	    sizeof (m_stat->m_machine) - 1);
 
-	m_stat->m_num_fpus = (uint_t)sysconf(_SC_NPROCESSORS_CONF);
-
 	if (1 == m_stat->m_reprobe) {
 		/* Reprobe request */
 		fpsd_message(FPSD_NO_EXIT, FPS_DEBUG, REPRBE_REQ);
 		fpsd.d_iteration = 0;
 		fpsd.d_interval = 0;
 		fpsd.d_fpuid_index = 0;
-		m_stat->m_num_fpus = 0;
 		m_stat->m_num_on_fpuids = 0;
 		m_stat->m_cpuids_size = 0;
 		m_stat->total_iter = 0;
@@ -441,10 +438,7 @@ fpsd_probe(mach_conf_t *m_stat)
 	 * cpu_info for each.
 	 */
 
-	m_stat->m_num_on_fpuids = (uint_t)sysconf(_SC_NPROCESSORS_ONLN);
-	fpsd_message(FPSD_NO_EXIT, FPS_DEBUG, NUM_ONLN_CPUS,
-	    m_stat->m_num_on_fpuids);
-	cpuid_list = (processorid_t *)malloc(m_stat->m_num_on_fpuids *
+	cpuid_list = (processorid_t *)malloc(m_stat->m_num_fpus *
 	    sizeof (processorid_t));
 	if (NULL == cpuid_list) {
 		fpsd_message(FPSD_NO_EXIT, FPS_INFO, LIBRARY_CALL_FAIL,
@@ -453,15 +447,18 @@ fpsd_probe(mach_conf_t *m_stat)
 	}
 
 	cpuid_index = 0;
-	for (i = 0; i < max_cpuids; i++) {
+	for (i = 0; i < m_stat->m_max_cpuid; i++) {
 		if (p_online(i, P_STATUS) == P_ONLINE) {
 			cpuid_list[cpuid_index++] = i;
 		}
-		if (cpuid_index == m_stat->m_num_on_fpuids) {
+		if (cpuid_index == total_onln) {
 			/* Break after all onln cpuids found */
 			break;
 		}
 	}
+	m_stat->m_num_on_fpuids = (uint_t)cpuid_index;
+	fpsd_message(FPSD_NO_EXIT, FPS_DEBUG, NUM_ONLN_CPUS,
+	    m_stat->m_num_on_fpuids);
 
 	/*
 	 * Get cpu-brand info all valid cpuids using kstat.
@@ -616,7 +613,7 @@ parse_and_set_cpu_id_list(char *strCPUs)
 			invalid = 1;
 		}
 		if (num_cpus == fpsd.d_conf->m_num_fpus) {
-			/* More than max configurable cpus */
+			/* More than max supported cpus */
 			fpsd_message(FPSD_NO_EXIT, FPS_ERROR,
 			    INVAL_PROP_VALUE, strCPUs);
 			invalid = 1;
@@ -808,23 +805,23 @@ static int fpsd_init() {
 
 	m_conf_p = fpsd.d_conf;
 	m_conf_p->m_machine[0] = '\0';
-	m_conf_p->m_num_fpus = 0;
 	m_conf_p->m_num_on_fpuids = 0;
 	m_conf_p->m_cpuids_size = 0;
 	m_conf_p->total_iter = 0;
 	m_conf_p->m_reprobe = 0;
 	m_conf_p->m_num_cpus_to_test = 0;
+	m_conf_p->m_num_fpus = (uint_t)sysconf(_SC_NPROCESSORS_MAX);
 
 	(void) mutex_init(&log_mutex, USYNC_THREAD, NULL);
 
-	max_cpuids = (int)sysconf(_SC_CPUID_MAX) + 1;
+	m_conf_p->m_max_cpuid = (int)sysconf(_SC_CPUID_MAX) + 1;
 
 	/*
 	 * Allocate enough memory to accomodate maximum number of CPUs
 	 * supported by this platform.
 	 */
 	m_conf_p->m_cpus = malloc(sizeof (fps_cpu_t) *
-			(int)sysconf(_SC_NPROCESSORS_MAX));
+			m_conf_p->m_num_fpus);
 	if (NULL == m_conf_p->m_cpus)
 		return (1);
 	else
diff --git a/usr/src/cmd/fps/fpsd/fpsd_sched.c b/usr/src/cmd/fps/fpsd/fpsd_sched.c
index a37deb4b2a..283feaca66 100644
--- a/usr/src/cmd/fps/fpsd/fpsd_sched.c
+++ b/usr/src/cmd/fps/fpsd/fpsd_sched.c
@@ -344,7 +344,7 @@ test_fpu_thr(/* ARGSUSED */ void *arg)
  */
 #define	MINSLEEP	8
 
-	num_cpus =  sysconf(_SC_NPROCESSORS_ONLN);
+	num_cpus = fpsd.d_conf->m_num_on_fpuids;
 
 	intvl =  poll_intvl = fpsd.d_interval;
 
@@ -586,6 +586,32 @@ test_fpu_thr(/* ARGSUSED */ void *arg)
 }
 
 /*
+ * get_num_onln_cpus(): returns the number of processors that are in
+ * "on-line" state only. This number will be less than the number
+ * returned by sysconf(_SC_NPROCESSORS_ONLN) if there are some
+ * processors in "no-intr" state.
+ */
+
+static int
+get_num_onln_cpus()
+{
+	int i;
+	int num_onln = 0;
+	int total_onln = sysconf(_SC_NPROCESSORS_ONLN);
+
+	for (i = 0; i < fpsd.d_conf->m_max_cpuid; i++) {
+		if (p_online(i, P_STATUS) == P_ONLINE) {
+			num_onln++;
+		}
+		if (num_onln == total_onln) {
+			/* Break after all onln cpuids found */
+			break;
+		}
+	}
+	return (num_onln);
+}
+
+/*
  * Identifies the fpu on which test will be scheduled next.
  */
 
@@ -603,7 +629,8 @@ identify_fpu_to_run_test(int *freq, int *iteration, int *fpu_index) {
 
 	*iteration = *freq = 0;
 	while (fpuid == -1) {
-		num_onln = (int)sysconf(_SC_NPROCESSORS_ONLN);
+		/* Check if the number of online processors has changed */
+		num_onln = get_num_onln_cpus();
 		if (num_onln != fpsd.d_conf->m_num_on_fpuids) {
 			fpsd_message(FPSD_NO_EXIT, FPS_DEBUG, REPROBE_MSG);
 			fpsd.d_conf->m_reprobe = 1;